Ukubuyela emuva kwamasiko ngokumelene ne-generative ai | nguStephanie Kirmer | Feb, 2025
Yini eyenza abantu abaningi bacasukela i-ai ekhiqizayo, futhi lokho kube naliphi ithonya ezinkampanini ezibhekene nazo?
Ukuvezwa kwakamuva kwe-Deepseek-R1, isikali esikhulu se-LLM esakhiwe yinkampani yaseChina (ne-Deepseek), kube ngumcimbi othakazelisa kakhulu walabo bethu abachitha isikhathi sokubuka nokuhlaziya amasiko nezenhlalo azungeze i-AI. Ubufakazi buveza ukuthi u-R1 waqeqeshwa ingxenye yentengo yentengo ebiza ukuqeqesha i-chatgpt (noma yimaphi amamodeli awo wakamuva, empeleni), futhi kunezizathu ezimbalwa ezingase zibe yiqiniso. Kepha lokho akusikho ngempela engifuna ukukhuluma ngakho lapha – amathani abalobi abacabangayo aphawule ngokuthi yikuphi okujulile-R1 okukhona, nokuthi kwenzekani ngempela kwinqubo yokuqeqeshwa.
Engikuthanda kakhulu okwamanje ukuthi lezi zindaba zishintshele kanjani ezinye zomfutho endaweni ye-AI. I-NVIDIA namanye amasheya ahlobene aphonswe ngokuqinile lapho kuvela izindaba ze-Deepseek-R1, ikakhulukazi (kubonakala sengathi (kubonakala sengathi (kubukeka sengathi ayidingi kahle i-GPUS ukuqeqesha, futhi ngokuziqeqesha kahle, kudinga amandla amancane kunemodeli ye-OpenAi. Sengivele ngicabanga nge-backlash yamasiko i-AI enkulu ebhekene nayo, futhi into enjengale evulwa isikhala esithe xaxa sabantu ukuba bagxeke imikhuba nezithembiso zezinkampani ze-AI generative ai.
Sikuphi ngokuya ngamazwi abucayi abhekene ne-ai generative ai njengebhizinisi noma njengobuchwepheshe? Kuvelaphi lokho kuvela, futhi kungani kwenzeka?
Lezi zingcindezi ezimbili ezivame ukugcwala ngokugqamisa engicabanga ukuthi zithakazelisa kakhulu okokuqala, umbono wezenhlalo noma womphakathi, okwesibili, umbono ongokoqobo. Ngokombono omuhle wezenhlalo, ukugxekwa kwe-ai ekhiqizayo njengebhizinisi kanye nomkhakha kuyiminyaka, futhi ngikhulume kakhulu ngabo ekubhaleni kwami lapha. Ukwenza i-AI ekhiqizayo ibe yinto ebukhali iza ngezindleko ezingejwayelekile, kusukela kwezemvelo kuya kwezomnotho nangale kwalokho.
Njengendaba engokoqobo, kungenzeka kube lula ukubilisa phansi ukuze “lobu buchwepheshe abusebenzi ngendlela esathenjiswa ngayo”. Ama-AI akhiqizayo alele kithi, noma “Hallucinates”, futhi enza kabi ezinhlotsheni eziningi zemisebenzi esidinga usizo kakhulu kosizo lwezobuchwepheshe ku-. Siholelwa ekukholelwe ukuthi singakwethemba lobu buchwepheshe, kepha sehluleka ukufeza okulindelwe, kanti ngasikhathi sinye sisetshenziselwa izinto ezihlukumezanayo ezingezinhle futhi zinobugebengu njenge-CSAM eyenziwe ngentando yeningi.
Ngakho-ke lapho sibheka lezi ndawonye, ungahlakulela impikiswano eqinile: lobu buchwepheshe abuphili kulokho okulindelwe okweqile, futhi ngokushintshana kwalesi simo sokusebenza, sinikela ngogesi, amanzi, isimo sezulu, imali, amasiko, kanye Nemisebenzi. Akuyona ukuhweba okusebenzayo, emehlweni abantu abaningi, ukubeka ngobumnene!
Ngithanda ukuletha i-nuance encane esikhaleni, ngoba ngicabanga ukuthi lapho semukela ukulinganiselwa kokuthi i-ani eyenziwe ngento eyenzekayo, nokulimala okungakubangela, futhi ungadlali umdlalo ophezulu, singathola indawo ephakathi nendawo . Angicabangi ukuthi kufanele sikhokhe inani elinamatshe okuqeqeshwa kanye nokuzithoba kwalezi zimodeli ngaphandle kokuthi imiphumela empeleni, ikufanele ngempela. Ukuhlakulela ama-molecule amasha ocwaningo lwezokwelapha? Mhlawumbe, yebo. Ukusiza izingane ukukopela (kabi) emsebenzini wesikole? Cha ngiyabonga. Angiqiniseki ukuthi kufanelekile izindleko zangaphandle ukungisiza ngibhale ikhodi kancane kahle emsebenzini, ngaphandle kokuthi ngenze okuthile okubaluleke kakhulu. Kudingeka sithembeke futhi sibhekane namaqiniso ngentengo yangempela yokwakha nokusebenzisa lobu buchwepheshe.
Ngakho-ke, nalokho kusho ukuthi, ngithanda ukungena ngingene ngibheke ukuthi lesi simo saba kanjani. Ngabhala emuva ngoSepthemba 2023 ukuthi ukufundwa kwemishini kunenkinga yokubona komphakathi, futhi endabeni ye-AI ekhiqizayo, ngicabanga ukuthi kufakazelwe yimicimbi. Ngokukhethekile, uma abantu bengenakho okulindelwe okungokoqobo nokuqonda ukuthi ama-LLM alungele ini nokuthi yini abangayilungele, bazophuma, bese behlehlela emuva.
“Ingxabano yami ihamba into efana nale:
1. Abantu abakulungeli ngokwemvelo ukuze baqonde futhi bahlanganyele nokufunda komshini.
2. Ngaphandle kokuqonda la mathuluzi, abanye abantu bangakugwema noma bangabethembi.
3. Okubi kakhulu, abanye abantu bangase basebenzise kabi la mathuluzi ngenxa yokuthola amanga, okuholela emiphumeleni eyingozi.
4. Ngemuva kokubhekana nemiphumela emibi yokusebenzisa kabi, abantu bangase babe manqikanqika ukwamukela amathuluzi okufunda womshini wesikhathi esizayo angathuthukisa izimpilo nemiphakathi yazo. “
mina, enkingeni yokuqonda komphakathi yokufunda komshini, sept 2023
Ngakho kwenzekani? Yebo, imboni ye-AI evamile ye-AI ietha i-DOBO IQHAZA EKUQALENI KANYE NOKUTHOLA IZIMPAWU.
Ingxenye yenkinga ukuthi i-ai ekhiqizayo ayikwazi ngempela ukwenza konke okushiwo yi-hype izimangalo. I-LLM ayikwazi ukusetshenziselwa ukuphendula imibuzo, ngoba akuyona “umshini wamaqiniso”. Yigama elilandelayo elilandelayo emshinini wemisho “. Kepha sibona izithembiso zazo zonke izinhlobo ezingakunaki lokhu kulinganiselwa, futhi izinkampani zezobuchwepheshe ziphoqa izici ze-AI ezikhiqizayo kulo lonke uhlobo lwesoftware ongacabanga khona. Abantu babezonda i-Clipppy ye-Microsoft ngoba bekungeyona into enhle futhi bengafuni ukuba nayo edonswe umphimbo wazo – futhi umuntu angasho ukuthi benza into efanayo eyisisekelo ngenguqulo ethuthukisiwe, futhi siyabona ukuthi abanye abantu basaqhubeka Kuyacasukelazwa.
Lapho othile eya e-LLM namuhla futhi acele inani lentengo yezithako eresiphi esitolo sazo sendawo esitolo manje, akukho ithuba lokuthi imodeli ingakuphendula kahle lokho, ngokuthembekile. Lokho akuphakathi kwamakhono ayo, ngoba imininingwane yeqiniso mayelana nalezo zintengo ayitholakali kwimodeli. Imodeli ingahle iphumele ngephutha ukuthi isikhwama sezaqathe yi- $ 1.99 ku-Publix, kodwa yilokho nje, ingozi. Ngokuzayo, ngamamodeli wokuhlanza kanyekanye emafomini amenti, kunethuba lokuthi singakha imodeli emincane ukwenza lolu hlobo lwento ngendlela efanele, kodwa manje kunguBogus ngokuphelele.
Kepha abantu bacela ama-LLMS le mibuzo namuhla! Futhi lapho befika esitolo, badumele kakhulu ngokuqanjwa amanga ngobuchwepheshe ababecabanga ukuthi yibhokisi lezimpendulo zomlingo. Uma u-OpenAI noma i-anthropic, ungahle unciphise, ngoba uma lowo muntu ekukhokhele imali ekhokhwa njalo ngenyanga, usuvele usunayo imali. Futhi uma bengazange, kahle, uthole inombolo yomsebenzisi ukukhulisa enye futhi, futhi lokho kukhula.
Kodwa-ke, empeleni lokhu kuyinkinga enkulu yebhizinisi. Lapho umkhiqizo wakho wehluleka kanjena, ngokusobala, obonakalayo (okungenakugwenywa (okungenakugwenywa (okungenakugwenywa!), Uqala ukubhukuda ibhuloho phakathi kwalowo msebenzisi nomkhiqizo wakho. Kungase kungakushisi konke ngasikhathi sinye, kepha kancane kancane kudiliza ubuhlobo umsebenzisi onawo ngomkhiqizo wakho, futhi uthola amathuba amaningi kakhulu ngaphambi kokuba othile anikeze umsebenzisi akhuphukele emkhayeni. Endabeni ye-generative ai, kubonakala sengathi awutholi amathuba amaningi. Futhi, ukwehluleka kwimodi eyodwa kungenza abantu bathembele lonke ubuchwepheshe kuzo zonke izinhlobo zalo. Ngabe lowo msebenzisi ezokwethemba noma ukukukholelwa eminyakeni embalwa lapho ubopha i-LLM i-Backhendezela ku-Realtime Prime APIs futhi empeleni ingabuyisa amanani esitolo okudla? Angiboni. Lowo msebenzisi kungenzeka angavumeli imodeli yakho isize ukubuyekeza ama-imeyili kubasebenza nabo abasebenza nabo ngemuva kokuyehluleka komunye umsebenzi.
Kulokho engikubonayo, izinkampani ze-tech zicabanga ukuthi zingavele zigqoke abantu phansi, zibephoqe ukuthi ziye zamukele ukuthi i-generative ai iyingxenye engenakugwemeka yayo yonke isoftware yabo manje, noma ngabe ayisebenzi. Mhlawumbe, kodwa ngicabanga ukuthi leli isu lokuzimela. Abasebenzisi bangase bahlangane futhi bamukele isimo sezinto, kepha ngeke bazizwe benekhono ngobuchwepheshe noma babheke emgabeni wakho. Ukwemukelwa nge-BeGrudging akuyona uhlobo lwamandla ofuna uhlobo lwakho lomkhiqizo ugqugquzele phakathi kwabasebenzisi!
Ungase ucabange, kahle, lokho kucacile ngokwanele – mali emuva ezicini ezikhiqizayo ze-AI ku-software, futhi zisebenzise nje kwimisebenzi lapho ingawuphawula umsebenzisi futhi isebenze kahle. Bazoba nesipiliyoni esihle, bese ubuchwepheshe buba ngcono, sizongeza okuningi lapho kunengqondo khona. Futhi lokhu bekuzoba ukucabanga okunengqondo (yize ngakusho njalo, izindleko zangaphandle ziphakeme kakhulu emhlabeni wethu nemiphakathi yethu).
Kodwa-ke, angicabangi ukuthi abadlali abakhulu be-AI abangenza bangakwenza ngempela lokho, futhi nakhu. Abaholi bezobuchwepheshe basebenzise imali ebaluleke kakhulu yemali ekwakheni nasekuzameni ukuthuthukisa lobu buchwepheshe – kusuka ekutshaleni imali ezinkampanini eziyizithuthukisa, ukwakha izikhungo zedatha, kunamakhulu ezigidigidi zamaRandi ashonile Kulesi sikhala sekuvele kuningi okuzayo.
Embonini yezobuchwepheshe, okulindelwe inzuzo kuhluke kakhulu kulokho ongabhekana nakho kweminye imikhakha – ukuqala kwesoftware ye-VC okumele kwenziwe kabusha (kuya ngesiteji) ukuze kubhekwe njengempumelelo ye-Standy. Ngakho-ke abatshalizimali ezinkampanini ze-tech push, ngokusobala noma ngokuphelele, ukuthatha ukushintshwa okukhulu nobungozi obukhulu ukuze kuphinde kube yibuyiseli eliphakeme. Lokhu kuqala ukukhula kulokho esikubiza ngokuthi “bubble” – amanani angaba wukuqondanisa namathuba angempela ezomnotho, akhuphuka aphezulu futhi aphakeme ngaphandle kwethemba lokuba nethemba lokuba yiqiniso. Njengoba uGerrit de Vynck eposini laseWashington abonile, “… Abahlaziyi bakaWall Street balindele izinkampani ezinkulu zobuchwepheshe ukuba basebenzise ama-Revol $ 60 billion ngonyaka ngemali engenayo evela e-AI ngalelo phuzu … Ama-capitasts we-venture athulule izigidigidi ezinkulungwaneni ezinkulungwaneni ze-AI Start. I-AI Boom isizile ukufaka isandla kuma- $ 55,6 billion abatshaliswayo bafaka ama-US Start-Ups ngekota yesibili ka-2024, inani eliphakeme kakhulu ngekota elilodwa eminyakeni emibili, ngokusho kwe-Venture Capital Firm Pitchbook. “
Ngakho-ke, unikezwe izigidigidi ezitshaliwe, kunezimpikiswano ezinkulu okufanele zenziwe ukuthi inani elitshaliwe ekwakheni i-AI lwanele kuze kube manje akunakwenzeka ukufanisa nembuyiselo. Akuyona imali eningi okufanele yenziwe lapha, ngalobu buchwepheshe, ngokuqinisekile ayiqhathaniswa nenani elitshaliwe. Kepha, izinkampani ngokuqinisekile zizozama. Ngikholwa ukuthi yingxenye yesizathu esenza sibone ukukhiqizwa kwe-AI kufakwe kuzo zonke izinhlobo zokusebenzisa amacala lapho kungahle kube lusizo olukhulu, luphumelele, noma lwamukelwe. Endleleni, “Sisebenzise yonke le mali kulobu buchwepheshe, ngakho-ke kufanele sithole indlela siyithengise ngayo” uluhlobo lohlaka. Khumbula futhi ukuthi ukutshalwa kwezimali kuyaqhubeka nokushiswa ukuzama ukwenza umsebenzi we-tech, kepha noma iyiphi intuthuko ye-LLM kulezi zinsuku ifakazela kakhulu futhi iyanda.
Amathuluzi we-AI akhiqizayo awahloseli abalulekile ezimpilweni zabantu, ngakho-ke i-Calculation Calcurus ayisebenzi ukwenza umkhiqizo utholakala futhi ukhombise abantu ukuze bayithenge. Ngakho-ke, sibona izinkampani zithuthela kumodeli “yesici” se-ai ekhiqizayo, engiziphendulayo engozini ngo-Agasti 2024 Ukwenza izici nentengo ehambisana nalokhu yandisa konke okuphoqelekile. Ngiyavuma ukuthi angizange ngixhumeke phakathi kwenkinga yezithombe zomphakathi kanye nenkinga yemodeli yomkhiqizo ye-vs vs kuze kube muva nje – kepha manje siyabona ukuthi ziphakathi. Ukunikeza abantu isici esinezinkinga zokusebenza esizibonayo, bese zibakhulisa ngenxa yaso, kuseyinkinga yangempela yezinkampani. Mhlawumbe lapho okuthile kungasebenzi ngomsebenzi, akuwona umkhiqizo noma isici? Uma lokho kuvela ukuthi kunjalo, abatshalizimali e-AI verative bazoba nenkinga yangempela ezandleni zabo, ngakho-ke izinkampani zizibophezela ekukhiqizeni izici ze-AI, noma ngabe ziyasebenza kahle noma cha.
Ngizobe ngibheke ngentshisekelo enkulu ukubona ukuthi izinto zithuthuka kanjani kulesi sikhala. Angilindeli ukuthi noma yikuphi ukugxuma okukhulu ekusebenzeni kwe-AI adressionative ai, yize kuye ngokuthi izinto zivela kanjani nge-Deepseek, singabona okuthile ngokusebenza kahle, okungenani ekuqeqeshweni. Uma izinkampani zilalela izikhalazo zazo zabasebenzisi kanye ne-pivot, ukukhomba i-ai ekhiqizayo kuzinhlelo zokusebenza empeleni zilusizo, zingaba nethuba elingcono lokuthola i-backlash, okungcono noma okubi. Kodwa-ke, lokho kimi kubonakala kakhulu, akunakwenzeka kakhulu ukuba ahambisane nezisusa zenzuzo yenzuzo ephelelwe yithemba ababhekene nazo. Endleleni, sizogcina sichitha izinsizakusebenza ezinkulu ekusetshenzisweni kobuwula kwe-AI generative ai, esikhundleni sokugxila emizamweni yethu ekuthuthukiseni ukusetshenziswa kwezobuchwepheshe okukufanele ukuhweba.
